Parallel dataset collection across multiple simulators

[JordanCoin]

Summary

Run multiple Claude Code sessions in parallel, each with its own iOS Simulator + ATL instance on a different port, to collect cart dataset samples faster.

Setup

Each session needs:

• Its own booted simulator (iPhone 16, iPhone 15, etc.)
• ATL app running on a unique port (9222, 9223, 9224...)
• A Claude Code ralph loop with --port <N> on every CLI command
• A subset of merchants to avoid duplicates

Example: 3 parallel sessions on i9

Session	Port	Simulator	Merchants
1	9222	iPhone 17 Pro	Amazon, Target, Walmart
2	9223	iPhone 16	Best Buy, eBay, Walgreens
3	9224	iPhone 15	Nike, Kohls, Macys, Nordstrom

Prerequisites

• Verify multiple simulators can run simultaneously on the i9
• Confirm ATL app can launch on custom ports (or run multiple instances)
• Test atl dataset append is safe for concurrent writes to the same metadata.jsonl
• Split collect-dataset.md prompt into per-session merchant lists

Estimated throughput

• 1 session: ~50 samples/day
• 3 sessions: ~150 samples/day
• Target: 200+ samples across 10 merchants in 1-2 days

Notes

Start with a single session first to validate the collection loop works end-to-end before scaling to parallel. The dataset dir (~/.atl/dataset/) should probably use file locking or separate dirs per session to avoid write conflicts.